OpenAI's realtime voice and video features in ChatGPT have advanced little since the GPT-4o release in May 2024, with realtime-v2 adding reasoning but often yielding repetitive outputs
Observers blame insufficient focus and cite Thinking Machines for faster gains
——0——
Agree there hasn't been that much progress* I think that's mostly just due to neglect rather than that one can't do better, e.g. Thinking Machines seems to have made rapid progress by actually working hard on it
*have only briefly tried realtime-v2, like the incorporation of reasoning but it gets repetitive in "holding pattern" moments
Timothy B. Lee@binarybits@Miles_Brundage An example I've been thinking about recently is that ChatGPT has barely improved on the realtime voice/video stuff they showed off in May 2024 with the release of GPT-4o.
10:54 AM · May 20, 2026 · 615 Views
5:37 PM · May 20, 2026 · 193 Views